Deep breathing couples CSF and venous flow dynamics

Venous system pathologies have increasingly been linked to clinically relevant disorders of CSF circulation whereas the exact coupling mechanisms still remain unknown. In this work, flow dynamics of both systems were studied using real-time phase-contrast flow MRI in 16 healthy subjects during normal and forced breathing. Flow evaluations in the aqueduct, at cervical level C3 and lumbar level L3 for both the CSF and venous fluid systems reveal temporal modulations by forced respiration. During normal breathing cardiac-related flow modulations prevailed, while forced breathing shifted the dominant frequency of both CSF and venous flow spectra towards the respiratory component and prompted a correlation between CSF and venous flow in the large vessels. The average of flow magnitude of CSF was increased during forced breathing at all spinal and intracranial positions. Venous flow in the large vessels of the upper body decreased and in the lower body increased during forced breathing. Deep respiration couples interdependent venous and brain fluid flow—most likely mediated by intrathoracic and intraabdominal pressure changes. Further insights into the driving forces of CSF and venous circulation and their correlation will facilitate our understanding how the venous system links to intracranial pressure regulation and of related forms of hydrocephalus.

prompts an upward surge of CSF from the lumbar region along the entire spinal canal, into the cranial vault and towards the subarachnoid and brain ventricular spaces 9 . During exhalation CSF was observed to move downward to varying extent. Cardiac-driven CSF pulsations, most prevalent in cervical regions in close proximity to the heart, are superimposed by a dominating, high-amplitude flow component triggered by forced respiration [9][10][11] . Furthermore, the inspiratory lowering of the intrathoracic pressure governs venous drainage out of the head/neck region as a prerequisite to ensure adequate preload of the heart. Real-time flow MRI demonstrated fluctuations of outflow in the cervical epidural veins dependent on the strength of inhalation 11 . These latest insights point to a tightly regulated equilibrium between CSF and venous systems, that might play an essential role in regulating the intracranial volume in accordance with the Monro-Kellie doctrine 12 .
The notion of a dynamic interplay between CSF and venous flow is in line with an increasingly popular holistic view on brain fluids 13,14 . Mounting clinical evidence substantiates the importance of the cerebral venous systems for maintaining intracranial pressure (ICP) and hence its role in a broad spectrum of neurological diseases like communicating hydrocephalus and myelopathy 15,16 . A high prevalence of extra-and intracranial venous flow obstruction was found in patients with idiopathic intracranial hypertension (IIH) 17 . However, the physiological mechanisms, which link CSF and veins, are still poorly understood.
Our study aimed (i) to provide further experimental support that respiration couples CSF and venous fluid systems, (ii) to include both normal free breathing and deep respiration in the study protocol, and (iii) to extend such measurements to the lower body.

Results
Flow dynamics during normal and forced breathing. Magnitude images possess a high sensitivity to through-plane flow because the inflow of previously unsaturated spins into the measurement section increases its signal intensity relative to the saturated spins within the slice. On the other hand, dark and bright signals in velocity maps correspond to opposite flow directions with grey values representing zero movement, i.e. stationary tissue. In all sections, the occurrence of bright signals in velocity maps refers to upward flow, while dark signals represent downward flow.
CSF and venous blood flow values over time at the Aqd as well as C3 and L3 for all 16 subjects are presented as a heat map in Fig. 2. Color-coded flow rates averaged across subjects are shown in Fig. 3 and flow rates averaged across time frames for the individual subjects are given in Suppl. Fig. 1. No flow signal could be detected in epidural veins at C3 (C3 EV) for subject #6 and in epidural veins at L3 (L3 EV) for subject #3.
As a general observation, CSF and venous blood flow reveal only minor respiration-driven dynamics during normal breathing. Forced respiration on the other hand, induced high-amplitude flow modulations at all locations in parallel to the breathing protocol (Figs. 2, 3). In the CSF system forced inspiration elicits an increase of flow in cranial direction (coded in red), i.e. at the Aqd, C3 and L3, whereas forced expiration causes a caudal CSF movement (coded in blue). The downward venous blood flow in the upper body (C3-EV, internal jugular veins at C3 (C3-IJV)) yields respiratory modulation with the onset of forced breathing. In the inferior vena cava (IVC) of the lower body the steady upward flow decreases during forced inspiration and rapidly increases again during forced expiration. Although flow in the EV at L3 shows great interindividual variability, respiratory modulations emerge during forced breathing parallel to those of the IVC. Figure 4 illustrates the temporal average of flow magnitude (mean across time frames) at all locations for each subject. The mean and SD across all subjects during normal and forced respiration and corresponding p-values of the paired t-tests are listed in Table 1. The average of the magnitude of CSF flow increases significantly during forced respiration at all locations (Aqd, C3, L3). Venous magnitude flow decreases significantly at C3-IJV and increases at L3-IVC with forced respiration, whereas no significant difference was found in EV (at C3, L3) compared to normal breathing.
Coupling of CSF and venous flow. The results of a correlation analysis between CSF and venous flow time series in both upper body and lower body are displayed in Fig. 5. Individual correlation coefficients for normal and forced breathing were calculated for C3-CSF v/s EV, C3-CSF v/s IJV, L3-CSF v/s EV, and L3-CSF v/s IVC. For all subjects and all locations (Fig. 5 right column) forced respiration tends to cause a stronger correlation between CSF and venous flow as compared to normal breathing. For most subjects flow during forced breathing at C3-CSF and L3-CSF is negatively correlated with C3-IJV and L3-IVC, respectively. In contrast, L3-CSF and L3-EV flow shows a positive correlation for most of the subjects (13/16). The paired t-tests reveal significant changes in correlation coefficients for normal and forced breathing, hinting at forced breathing to synchronize CSF and venous flow in the large vessels. The change for CSF v/s EV is not significant at both locations.
Respiratory and cardiac contributions to flow spectrum (frequency analysis). Figure 6 depicts the frequency spectra of the flow dynamics of each subject in all measured regions of the venous and CSF system during normal (left column) and forced breathing (middle column). During normal breathing, the maximum frequency components for CSF and venous flow at C3 are predominantly located between 0.9 and 1.5 Hz and thus in the range of heart rates with corresponding R/C values < 1 (right column) which summarize these findings by the ratio of low (respiratory) and high (cardiac) frequency components. At L3, dominant frequencies of < 0.5 Hz correspond to respiratory modulations for CSF as well as for EV flow resulting in R/C values > 1. In     www.nature.com/scientificreports/ the Aqueduct, no predominance of cardiac or respiratory modulations is found and hence the R/C during normal breathing is close to 1. During forced breathing the prevailing frequency component shifted to 0.2 Hz which matches the periodicity of the breathing protocol. The corresponding R/C values were > 1 at all locations except for C3-CSF where it remained < 1. The increase of the R/C value from normal to forced breathing is significant in all positions (Fig. 6).
ROI sizes during normal and forced breathing. The temporal evolution of the individual ROI sizes (mm 2 ) for all 16 subjects is depicted in Suppl. Fig. 2. ROI areas in C3-IJV and L3-IVC slightly fluctuate in parallel to forced breathing. In 6 of the 7 ROIs, no consistent temporal alterations were observed during normal or forced respiration and the comparison between both parts of the protocol reveals no significant changes (Suppl.

Discussion
The central observation of this work is the fact that the dynamics of CSF and venous blood flow vary distinctly between normal and forced breathing and thus confirm the strong dependence of both systems on respiration.
In contrast to normal breathing where cardiac driven frequency components prevail, the beginning of deep respiration causes immediate adherence of fluid dynamics to that driving force. As demonstrated by the shift of the main frequency component of CSF flow, the transition from normal to forced breathing promptly causes a switch from cardiac-to respiration-related dynamics in all positions except for C3-CSF (Fig. 6). Moreover, forced inspiration elicits a distinct upward surge of CSF from the lumbar region up to the intracranial aqueduct in line with previous findings. In contrast, forced expiration leads to reversed, downward flow. Venous outflow from the head/neck region is also modulated by forced respiration which so far has only been demonstrated for the cervical epidural veins 11,18 . Simultaneously, the steady venous flow from the lower body up towards the heart abates at the onset of forced inspiration and rises quickly afterwards throughout expiration. CSF flow shows rapid respiration-dependent adjustments. Upward surge of CSF during forced inspiration is considered to balance the simultaneously enhanced venous pooling out of the head/neck region which is driven by lowering of the intrathoracic pressure in accordance with the Monro-Kellie doctrine 12 . Transmissions of intrathoracic and also intraabdominal pressure changes into the bony spinal canal are assumed to be facilitated by the presence of abundant connections from the extraspinal paravertebral venous plexus through the intervertebral foramina to the epidural spaces 19,20 . As a result, the high-capacity, valveless epidural venous plexus have been found to be kept open by negative pleural and epidural transmural pressures 14,19,21,22 . However invasive in vivo measurements of intrathoracic or intraabdominal pressures were not obtained in this study.
The amplified CSF movements during forced breathing resulted in a significant increase of average magnitude flow at all positions compared to normal breathing. As no significant variations in CSF ROI sizes occurred over time the rise of flow rates is ascribed to changes of flow velocities (Suppl. Table 1, Suppl. Fig. 2). Moreover, the shift of frequency components towards respiratory rates leads to a significant increase of R/C values thus corroborating deep breathing as the prevailing driving force.
In support of our findings, contributions of deep breathing to CSF velocities and flow rates have been shown in several previous real-time flow MRI studies albeit mainly at the aqueduct and foramen magnum. For the aqueduct Takizawa et al. 23 reported a greater amount of CSF displacement by respiratory than cardiac contributions, however, at lower velocities. A recent computational study conducted by Vinje et al. 24 demonstrated that small respiratory-induced pulsations of the ICP gradient induce CSF flow volumes dominating the cardiac component. Also Yildiz et al. 25 reported in single slice real-time flow MRI measurements at the foramen magnum smaller R/C ratios during natural breathing compared to deep breathing.
Venous flow in the upper body was measured in the IJV and cervical EV as well as in the IVC and lumbar EV of the lower body's venous system. The constant venous drainage from the head/neck region through IJV and EV alters distinctly with forced breathing and flow adheres to respiratory-driven patterns in agreement with previous studies 11,26 . The start of forced inspiration perturbs the steady venous return which prevails in the IVC during normal breathing. Instantaneously, flow ceases and even briefly reverses downward. During forced respiration the frequency of the main venous flow component adhered to the breathing protocol.
The average of venous flow magnitude showed a trend towards higher numbers in cervical and lumbar EV and IVC reaching significance only in the latter (Table 1). At the cervical level, we observed respiratory modulations in both IJV and EV as described by Epstein et al. 27 . While in the EV, flow only tended to increase over 40 s forced breathing, averaged flow magnitude in the IJV (Table 1) and their ROI areas decreased significantly (Suppl. Table 1). To our knowledge real-time cervical venous flow has not been studied during forced breathing in supine position so far. One of the reasons for the discrepancy between outflow through the IJV and EV could be hypocapnia as a probable hyperventilatory effect. As one limitation, we did not control CO 2 partial pressures, but explained signs of hyperventilation to the subjects and instructed them specifically to avoid these. Furthermore, hypocapnia affects cardiac output, arterial flow and to lesser extent cerebral blood volume [28][29][30] . Hence, its influence on the venous system might only be secondary and would be expected to occur in all veins. Flow differences between these two cerebrovenous drainage pathways have so far been found in relation to postural changes. Multiple studies assigned the dominance of IJV drainage to the supine position and pointed to the EV systems as the major pathway in the upright position 31 . The bony spinal canal protects the thin valveless venous plexus from direct exposure to the thoracic and abdominal pressure changes and constant negative pleural and epidural pressures as well as bridging trabeculae within their lumen prevent the collapse of the vessels 14,19,20,31,32 . In contrast, forced breathing might expose the IJV to various pressure changes along its course in the neck and at the entrance to the thoracic cavity.
Venous flow in the lower body was explored in the IVC. The steep decrease and reversal of flow in that vessel during forced breathing resulted in significantly increased average flow magnitudes (Table 1)  www.nature.com/scientificreports/ flow in L3-EV is particularly high during normal breathing, e.g. see Fig. 2. Individual venous plexus anatomy as well as factors such as breathing performance might play important roles. The results of 15/16 healthy subjects reported here represent one of the largest cohorts for flow measurements under physiological conditions. A trend towards higher average flow magnitudes was seen during forced breathing not reaching significance (Table 1). Their dynamics parallel the behavior in the IVC with upward movements during expiration. In recent years, clinically relevant disorders of CSF circulation and venous systems pathologies have increasingly been linked. The exact mechanisms which couple cerebral blood flow and CSF movement are up to this point not well understood. In our study, flow in the large veins yields a significantly negative correlation with corresponding CSF dynamics during forced breathing which is presumed to prompt sufficient intrathoracic and abdominal pressure changes (Fig. 5). Correlation coefficients showed a negative trend in C3-CSF v/s EV and a positive in L3-CSF v/s EV without reaching significance. This might be caused by a delayed response of flow in the EV to forced breathing or by a higher uncertainty of flow measurements due to the delicate vessel sizes and thus smaller amount of flow. Furthermore, pressure conditions in the epidural space have been demonstrated to vary along the spine which could result in less uniform flow behavior 35 .
Taken together, this study provides in vivo evidence that deep breathing represents a physiological coupling of cranio-spinal fluid flow. Although the clinical impact warrants further investigations, an increasing number of clinical observations point to an important pathophysiological role. For example, compromised venous flow is considered one cause of IIH, where increased intracranial pressure leads to headaches and papilledema 36,37 . Intracranial obstruction of venous outflow i.e. due to transverse sinus stenosis or extracranial IJV obstruction have been accounted for perturbed intracranial pressure regulation and CSF circulation 36,38 . Moreover, cerebral venous hypertension as a consequence of sleep-disordered breathing is discussed as one of the key factors for the development of normal pressure hydrocephalus, leading to ventriculomegaly and dementia 39 . In infants, Shulman et al. 15 related elevated pressure in the sagittal sinus to the occurrence of hydrocephalus. Thus, disturbances of the venous-CSF coupling mechanism might indeed represent a key in the pathophysiology of related CSF disorders.

Conclusions
In contrast to normal breathing, forced respiration leads to pronounced increases and a prompt synchronization of flow dynamics in CSF and venous systems. The lumbar caval and epidural venous upward surge during free breathing abates with the onset of forced inspiration and rises during ensuing exhalation-a pattern opposite to that of venous flow in the upper body part. Spinal and intracranial CSF, on the other hand, move uniformly upwards during forced inhalation as previously described. Our results provide evidence that deep respiration couples interdependent venous and brain fluid flow-most likely mediated by intrathoracic and intraabdominal pressure changes. Further insights into the tight interplay between CSF and venous fluid dynamics will expand our understanding of the pathophysiology of human diseases with CSF flow disturbances such as hydrocephalus and facilitate the development of more specific therapeutic options.

Methods
Subjects. Sixteen healthy volunteers (6 females, 10 males, age, 28 ± 5; (mean ± SD); height 178 ± 8 cm, weight 73 ± 12 kg, BMI 23 ± 2.4 kg/m 2 ) without contraindication for MRI and no known illness were enrolled. The study was approved by the institutional review board of the University Medical Center Göttingen (#18/2/14) and written informed consent was obtained from each subject prior to MRI. The study was in compliance with the Declaration of Helsinki. Study design. Real-time flow MRI was performed in three transversal cross-sections covering intracranial and spinal CSF and the venous systems at the Aqd, C3 and L3. ROIs for the analysis of CSF dynamics were placed in the Aqd and spinal subarachnoid spaces at C3 and L3 as outlined in Fig. 1a-e. For corresponding venous flow determinations ROIs were drawn around the IJVs (Fig. 1c) with the stronger flow signal in the neck region and around the lumbar IVC (Fig. 1e). The venous plexus expanding in the epidural spaces inside the entire vertebral column (internal vertebral venous plexus) commonly forms prominent orthogonal veins at C3 which resemble a rope ladder. The more prominent EV was selected for the ROI analysis (Fig. 1b). The lumbar venous plexus forms a spacious mesh rendering epidural vessels less well identifiable. ROIs in that region were defined around flow signals detectable in the epidural space ventral to the CSF space (Fig. 1d).
All subjects were examined in supine position and required to follow a visually presented breathing protocol. Timing and commands for deep inspiration and expiration were explained and trained beforehand as inspiration and exhalation had to occur gradually over 2.5 s periods. Furthermore, subjects were instructed to avoid hyperventilation. The breathing protocol started with 40 s of normal free breathing followed by 8 cycles of 2.5 s forced inspiration and 2.5 s forced expiration. The protocol concluded with another 10 s period of normal breathing summing up to a total of 90 s per scan (see Fig. 3, bottom line). Subjects were without any cardiac conditions, had pulse rates within normal age limits, while respiration was monitored via a belt fixed at the level of the diaphragm. Individual breathing performance and adherence to the protocol were evaluated by visual observation and measurements were repeated if deemed necessary. The flow data were acquired in the same order for all subjects and an individual study lasted about 40-50 min.
Real-time phase-contrast flow MRI. All data were acquired on a 3 Tesla scanner (Magnetom Prisma Fit, Siemens Healthcare) using real-time phase-contrast flow MRI based on highly undersampled radial FLASH sequences 40,41 with timing-optimized gradient design 42 . Quantitative velocity maps were obtained by a modelbased reconstruction technique offering access to high spatiotemporal resolutions 43 44 designed to accomplish automatic segmentation of flow signals in real-time image series after manual definition of a single initial ROI, which then serves as a seeding image for dynamic segmentation. The determination of through-plane flow was based on both signal intensities in magnitude images and corresponding phase difference values in velocity maps (see Fig. 1, right columns for representative examples). Further data processing was performed using Matlab (Mathworks, Massachusetts, USA). For a comparison of the two parts of the breathing protocol, flow rates and ROI sizes were presented as mean values averaged across time frames and grouped into normal (40 s) and forced (40 s) breathing. Prior to temporal averaging, the magnitude of each flow time series was taken to remove the flow's directionality. Further statistical analysis then was performed on the averaged flow magnitude. After a Shapiro Wilk test to test for normal distribution across subjects, a paired t-test was employed to find differences in averaged flow magnitude rates and ROI sizes between normal and forced respiration. Alternatively, a Wilcoxon Signed Rank test was applied for not normally distributed data. Statistical significance was accepted at p values < 0.05.
Temporal synchronisation of CSF and venous flow was studied using Pearson's correlations for each subject individually, to investigate whether CSF and venous system are coupled. Correlation coefficient pairs (from normal and forced breathing) of each subject were then used for a paired t-test to investigate the effect of respiration on synchronisation of CSF and venous flow.
To illustrate flow components related to respiration and cardiac pulsation, the time series data were analysed in the frequency domain. A power ratio analysis as described by Yildiz et al. determined the relative contribution of respiration versus cardiac (R/C) components of the flow data 25 . Here, R is defined as the area under curve of low frequencies (0-0.5 Hz) and C bases on a 0.5 Hz interval around the individual maximum frequency close to 1 Hz. R/C values larger than one thus indicate flow dynamics predominantly modulated by respiration and R/C values smaller than one a predominance of cardiac modulations. A paired t-test was applied to evaluate differences between forced and normal breathing.
Ethics approval and consent to participate. The study was approved by the institutional review board of the University Medicine Göttingen ((#18/2/14) and written informed consent was obtained from each subject prior to MRI. The study was in compliance with the Declaration of Helsinki.

Data availability
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.